torch.compile

torch.compile compiles PyTorch code to fused kernels to make it run faster. For training, it traces both the forward and backward pass together and compiles them into optimized kernels, reducing the overhead of individual op launches and fusing operations to cut memory bandwidth usage.

Set torch_compile=True in TrainingArguments to enable it. Training compiles both the forward and backward pass, unlike inference which only compiles the forward pass. Compilation happens on the first training step, so expect it to be significantly slower than subsequent steps.

from transformers import TrainingArguments

args = TrainingArguments(
    ...,
    torch_compile=True,
    torch_compile_backend="inductor",
    torch_compile_mode="reduce-overhead",
)

Backend

The default backend is inductor, which compiles to Triton kernels with AOTAutograd. This is the right choice for most training workloads. Use cudagraphs for fixed-shape inputs, or ipex for Intel CPU training.

Compile mode

Use the table below to help select a torch.compile mode.

Next steps

• See the torch.compile for inference guide for details on fullgraph compilation and inference benchmarks.